From: Lassi Kortela <xxxxxx@lassi.io>
Date: Friday, July 10, 2020 5:45 PM

>> At this point with the fate of SRFI 199 in the air, and in any case
>> my [Harold's] baby, the steps needed are adding you [John] and me
>> to the HTML documents [copyrights] for SRFI 170 and 198. You should
>> do the former, you and Lassi need to agree who'll be doing what
>> with the expanding 198; I added Lassi to the recipients. I have a
>> Chibi Scheme implementation for the current 198, but it might as
>> well wait for a pull until the new 198 firms up.
>
> Thanks. I can write the next draft of 198 unless Harold wants to do
> the honors.

Don't care about the honor, but everything from SRFI 170 to my own
current personal project has been hanging fire on SRFI 198 for a full
two weeks.  Since I have a lot of time available, I'll go ahead and
make a new draft based on your proposal, with the modifications we
previously discussed.

For naming, I believe nothing works better than Foreign Interface Error
Handling for the SRFI, thus for the slot accessor naming, and while
I'm at it I'll list them again with explanations:

(foreign-error:error-set ferr) → symbol

This partitions sets of errors like POSIX errnos vs. a particular
database's, or some random library that's being wrapped, etc. (which
come to think of it could also return POSIX errors, like if a file
can't be found), and the following items will by convention ideally
be populated in the same way for a given error set.

(foreign-error:code ferr) → alist

As previously mentioned, this class of data can't be reduced to a
specific set of slots.  For example, POSIX has errno number + errno
define name, but the databases I'm most familiar with have just a
numeric or alphanumeric code.  The libsodium cryptography library
has neither, the closest is the name of the function that failed,
which would properly be put in foreign-error:foreign-interface:

(foreign-error:message ferr [locale]) → string
(foreign-error:scheme-procedure ferr) → symbol
(foreign-error:foreign-interface ferr) → symbol

The above is where the SRFI naming choice affects the API, and is made
general, could be 'chown for POSIX, could be 'prepare-statement for
lots of database interfaces including the PostgreSQL over the wire
one, could be 'ecb-cipher-file for my backup program shelling out to
a small C program that uses libsodium.

(foreign-error:data ferr) → alist

I'll include notes that it's fine to use this SRFI in Scheme land for
checking (like types) before you burden the foreign interface with a
malformed query, and encouraging people to include 'arguments as one
of the items in the alist.

> I'll also join back in to help finish 170 ASAP. Sorry about the
> intermittent and delayed comments on it.

Not sure what's left to do that's *required*, other than:

* serious progress if not finalization of this error SRFI 198
* Incorporating the resulting changes due to it in the draft
* Ditto mechanically changing my Chibi Scheme sample
  implementation to use it for both POSIX errors and sanity
  checking before calling POSIX functions
* Updating or changing the copyrights to both 170 and 198

I've got some minor touch ups to the Chibi Scheme sample
implementation, and inserting a big caveat about it not properly
handling errno setting and getting, which will keep it a sample
vs. production implementation.

There's two things missing from it, one involves getting more into the
guts of Chibi Scheme's port system to add the optional buffer-mode
argument to fdes->[textual|binary]-[input|output]-port, I'm not sure
that's really important since it's so system specific.  That is, it
doesn't necessarily illuminate anything about interfacing to POSIX,
the major purpose of the sample implementation.

But there's one bit of pure Scheme which is currently "left as an
exercise to the reader", call-with-temporary-filename.  There would be
real value in finishing that useful helper procedure, since it can be
copied wholesale by other R7RS implementations.

- Harold